End of training

README.md

@@ -41,7 +41,7 @@ This model was trained with GRPO, a method introduced in [DeepSeekMath: Pushing
 - TRL: 0.15.2
 - Transformers: 4.51.3
 - Pytorch: 2.6.0
-- Datasets: 3.5.1
+- Datasets: 3.6.0
 - Tokenizers: 0.21.1
 
 ## Citations

adapter_config.json

@@ -24,13 +24,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "v_proj",
-    "o_proj",
-    "q_proj",
     "down_proj",
     "up_proj",
+    "k_proj",
+    "v_proj",
+    "q_proj",
     "gate_proj",
-    "k_proj"
+    "o_proj"
   ],
   "task_type": "CAUSAL_LM",
   "trainable_token_indices": null,

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1b06d57c60e1f7f54ad1258d16af6eebc53214324b7f015d2565aeb42838181b
+oid sha256:99c4ea0e40aef54a7b118b6a6c47f53c3a8df09d82a539d466dbf8dd1d7c4b6c
 size 73911112

all_results.json

@@ -1,8 +1,8 @@
 {
     "total_flos": 0.0,
-    "train_loss": 7.424532896038727e-07,
-    "train_runtime": 344.1703,
-    "train_samples": 14,
-    "train_samples_per_second": 0.93,
-    "train_steps_per_second": 0.058
+    "train_loss": 1.0554939626672422e-06,
+    "train_runtime": 1935.7748,
+    "train_samples": 20,
+    "train_samples_per_second": 0.207,
+    "train_steps_per_second": 0.013
 }

train_results.json

@@ -1,8 +1,8 @@
 {
     "total_flos": 0.0,
-    "train_loss": 7.424532896038727e-07,
-    "train_runtime": 344.1703,
-    "train_samples": 14,
-    "train_samples_per_second": 0.93,
-    "train_steps_per_second": 0.058
+    "train_loss": 1.0554939626672422e-06,
+    "train_runtime": 1935.7748,
+    "train_samples": 20,
+    "train_samples_per_second": 0.207,
+    "train_steps_per_second": 0.013
 }

trainer_state.json

@@ -4,215 +4,265 @@
   "best_model_checkpoint": null,
   "epoch": 5.0,
   "eval_steps": 500,
-  "global_step": 20,
+  "global_step": 25,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
-      "completion_length": 309.25,
-      "epoch": 0.5714285714285714,
-      "grad_norm": 0.5461550951004028,
-      "kl": 0.0009220477732014842,
+      "completion_length": 353.03125,
+      "epoch": 0.4,
+      "grad_norm": 0.46145281195640564,
+      "kl": 0.0008386401459574699,
       "learning_rate": 5e-07,
       "loss": 0.0,
-      "reward": 3.0049657989293337,
-      "reward_std": 3.3861664421856403,
-      "rewards/concensus_correctness_reward_func": 1.4184375147451647,
-      "rewards/consensus_reward_func": 0.375,
+      "reward": 1.3044774606823921,
+      "reward_std": 1.3216868042945862,
+      "rewards/concensus_correctness_reward_func": 0.18774999678134918,
+      "rewards/consensus_reward_func": 0.4375,
       "rewards/cumulative_reward_2": 0.0,
-      "rewards/final_correctness_reward_func": 0.4375,
-      "rewards/question_recreation_reward_func": 0.5458408743143082,
+      "rewards/final_correctness_reward_func": 0.1875,
+      "rewards/question_recreation_reward_func": 0.4195087403059006,
       "rewards/soft_format_reward_func": 0.0,
       "rewards/strict_format_reward_func": 0.0,
-      "rewards/xmlcount_reward_func": 0.22818750143051147,
+      "rewards/xmlcount_reward_func": 0.07221875246614218,
       "step": 2
     },
     {
-      "completion_length": 309.3333333333333,
-      "epoch": 1.0,
-      "grad_norm": 0.4232158064842224,
-      "kl": 0.0007590926543343812,
-      "learning_rate": 4.864543104251586e-07,
+      "completion_length": 352.125,
+      "epoch": 0.8,
+      "grad_norm": 0.5013545155525208,
+      "kl": 0.0008077259626588784,
+      "learning_rate": 4.91481456572267e-07,
       "loss": 0.0,
-      "reward": 2.918933798869451,
-      "reward_std": 2.6660856008529663,
-      "rewards/concensus_correctness_reward_func": 0.9195000032583872,
-      "rewards/consensus_reward_func": 0.25,
+      "reward": 1.8831986337900162,
+      "reward_std": 1.584916140884161,
+      "rewards/concensus_correctness_reward_func": 0.3190000057220459,
+      "rewards/consensus_reward_func": 0.375,
       "rewards/cumulative_reward_2": 0.0,
-      "rewards/final_correctness_reward_func": 0.9166666666666666,
-      "rewards/question_recreation_reward_func": 0.5955170492331187,
+      "rewards/final_correctness_reward_func": 0.4375,
+      "rewards/question_recreation_reward_func": 0.5861985795199871,
       "rewards/soft_format_reward_func": 0.0,
       "rewards/strict_format_reward_func": 0.0,
-      "rewards/xmlcount_reward_func": 0.23724999030431113,
+      "rewards/xmlcount_reward_func": 0.165500002913177,
       "step": 4
     },
     {
-      "completion_length": 332.90625,
-      "epoch": 1.5714285714285714,
-      "grad_norm": 0.49632948637008667,
-      "kl": 0.0008320739943883382,
-      "learning_rate": 4.472851273490984e-07,
+      "completion_length": 275.28125,
+      "epoch": 1.2,
+      "grad_norm": 0.5912909507751465,
+      "kl": 0.0009123186537181027,
+      "learning_rate": 4.6650635094610966e-07,
       "loss": 0.0,
-      "reward": 3.5371958827599883,
-      "reward_std": 3.318536162376404,
-      "rewards/concensus_correctness_reward_func": 1.9525625724927522,
-      "rewards/consensus_reward_func": 0.375,
+      "reward": 1.5717561542987823,
+      "reward_std": 1.3975044712424278,
+      "rewards/concensus_correctness_reward_func": 0.0546875,
+      "rewards/consensus_reward_func": 0.4375,
       "rewards/cumulative_reward_2": 0.0,
-      "rewards/final_correctness_reward_func": 0.5,
-      "rewards/question_recreation_reward_func": 0.5218521282076836,
-      "rewards/soft_format_reward_func": 0.0,
-      "rewards/strict_format_reward_func": 0.0,
-      "rewards/xmlcount_reward_func": 0.18778124800883234,
+      "rewards/final_correctness_reward_func": 0.25,
+      "rewards/question_recreation_reward_func": 0.5156624000519514,
+      "rewards/soft_format_reward_func": 0.015625,
+      "rewards/strict_format_reward_func": 0.015625,
+      "rewards/xmlcount_reward_func": 0.2826562523841858,
       "step": 6
     },
     {
-      "completion_length": 281.6666666666667,
-      "epoch": 2.0,
-      "grad_norm": 0.43755120038986206,
-      "kl": 0.0006982047537652155,
-      "learning_rate": 3.867370395306068e-07,
+      "completion_length": 289.4375,
+      "epoch": 1.6,
+      "grad_norm": 0.7356537580490112,
+      "kl": 0.0009301297905039974,
+      "learning_rate": 4.2677669529663686e-07,
       "loss": 0.0,
-      "reward": 9.014580527941385,
-      "reward_std": 6.108865926663081,
-      "rewards/concensus_correctness_reward_func": 6.691999999806285,
-      "rewards/consensus_reward_func": 0.75,
+      "reward": 2.088467765599489,
+      "reward_std": 2.1497373655438423,
+      "rewards/concensus_correctness_reward_func": 0.6559999994933605,
+      "rewards/consensus_reward_func": 0.1875,
       "rewards/cumulative_reward_2": 0.0,
-      "rewards/final_correctness_reward_func": 0.6666666666666666,
-      "rewards/question_recreation_reward_func": 0.4938725878794988,
+      "rewards/final_correctness_reward_func": 0.375,
+      "rewards/question_recreation_reward_func": 0.4805927164852619,
       "rewards/soft_format_reward_func": 0.0,
-      "rewards/strict_format_reward_func": 0.0,
-      "rewards/xmlcount_reward_func": 0.41204166164000827,
+      "rewards/strict_format_reward_func": 0.015625,
+      "rewards/xmlcount_reward_func": 0.37375001423060894,
       "step": 8
     },
     {
-      "completion_length": 303.46875,
-      "epoch": 2.571428571428571,
-      "grad_norm": 0.6612632870674133,
-      "kl": 0.0009638653064030223,
-      "learning_rate": 3.1137137178519977e-07,
+      "completion_length": 339.125,
+      "epoch": 2.0,
+      "grad_norm": 5.122199535369873,
+      "kl": 0.0009551418406772427,
+      "learning_rate": 3.75e-07,
       "loss": 0.0,
-      "reward": 2.8641053661704063,
-      "reward_std": 3.624991323798895,
-      "rewards/concensus_correctness_reward_func": 1.875,
-      "rewards/consensus_reward_func": 0.1875,
+      "reward": 1.356352262198925,
+      "reward_std": 0.8040661737322807,
+      "rewards/concensus_correctness_reward_func": 0.031312499195337296,
+      "rewards/consensus_reward_func": 0.125,
       "rewards/cumulative_reward_2": 0.0,
-      "rewards/final_correctness_reward_func": 0.125,
-      "rewards/question_recreation_reward_func": 0.4982929490506649,
+      "rewards/final_correctness_reward_func": 0.5,
+      "rewards/question_recreation_reward_func": 0.4886647742241621,
       "rewards/soft_format_reward_func": 0.0,
       "rewards/strict_format_reward_func": 0.0,
-      "rewards/xmlcount_reward_func": 0.17831249348819256,
+      "rewards/xmlcount_reward_func": 0.21137499436736107,
       "step": 10
     },
     {
-      "completion_length": 310.625,
-      "epoch": 3.0,
-      "grad_norm": 0.4002675414085388,
-      "kl": 0.0008224714838434011,
-      "learning_rate": 2.2935516363191693e-07,
+      "completion_length": 331.59375,
+      "epoch": 2.4,
+      "grad_norm": 0.6923710107803345,
+      "kl": 0.0008254477070295252,
+      "learning_rate": 3.147047612756302e-07,
       "loss": 0.0,
-      "reward": 2.8802956144014993,
-      "reward_std": 2.705232401688894,
-      "rewards/concensus_correctness_reward_func": 0.9346666640291611,
-      "rewards/consensus_reward_func": 0.3333333333333333,
+      "reward": 2.2641742639243603,
+      "reward_std": 2.478426046669483,
+      "rewards/concensus_correctness_reward_func": 0.7421875,
+      "rewards/consensus_reward_func": 0.25,
       "rewards/cumulative_reward_2": 0.0,
-      "rewards/final_correctness_reward_func": 0.9166666666666666,
-      "rewards/question_recreation_reward_func": 0.41537899772326153,
+      "rewards/final_correctness_reward_func": 0.5625,
+      "rewards/question_recreation_reward_func": 0.4648617208003998,
       "rewards/soft_format_reward_func": 0.0,
       "rewards/strict_format_reward_func": 0.0,
-      "rewards/xmlcount_reward_func": 0.28025000045696896,
+      "rewards/xmlcount_reward_func": 0.2446250095963478,
       "step": 12
     },
     {
-      "completion_length": 308.96875,
-      "epoch": 3.571428571428571,
-      "grad_norm": 0.5559430718421936,
-      "kl": 0.0008217594295274466,
-      "learning_rate": 1.4957614383675767e-07,
+      "completion_length": 269.21875,
+      "epoch": 2.8,
+      "grad_norm": 0.93145751953125,
+      "kl": 0.0008568289995309897,
+      "learning_rate": 2.5e-07,
       "loss": 0.0,
-      "reward": 5.252937890589237,
-      "reward_std": 4.048280119895935,
-      "rewards/concensus_correctness_reward_func": 3.2305000035557896,
-      "rewards/consensus_reward_func": 0.6875,
+      "reward": 1.4057147353887558,
+      "reward_std": 0.8676463589072227,
+      "rewards/concensus_correctness_reward_func": 0.10481250286102295,
+      "rewards/consensus_reward_func": 0.125,
       "rewards/cumulative_reward_2": 0.0,
       "rewards/final_correctness_reward_func": 0.4375,
-      "rewards/question_recreation_reward_func": 0.6550942175090313,
+      "rewards/question_recreation_reward_func": 0.45208977349102497,
       "rewards/soft_format_reward_func": 0.0,
-      "rewards/strict_format_reward_func": 0.0,
-      "rewards/xmlcount_reward_func": 0.2423437489196658,
+      "rewards/strict_format_reward_func": 0.015625,
+      "rewards/xmlcount_reward_func": 0.2706875056028366,
       "step": 14
     },
     {
-      "completion_length": 341.7916666666667,
-      "epoch": 4.0,
-      "grad_norm": 0.340472012758255,
-      "kl": 0.0008785632477762798,
-      "learning_rate": 8.067960709356478e-08,
+      "completion_length": 283.96875,
+      "epoch": 3.2,
+      "grad_norm": 29.920019149780273,
+      "kl": 0.003008850064361468,
+      "learning_rate": 1.8529523872436977e-07,
       "loss": 0.0,
-      "reward": 3.29603590319554,
-      "reward_std": 4.0770609776179,
-      "rewards/concensus_correctness_reward_func": 1.7687500171984236,
-      "rewards/consensus_reward_func": 0.3333333333333333,
+      "reward": 1.2449401542544365,
+      "reward_std": 0.9733890295028687,
+      "rewards/concensus_correctness_reward_func": 0.1639375016093254,
+      "rewards/consensus_reward_func": 0.1875,
       "rewards/cumulative_reward_2": 0.0,
-      "rewards/final_correctness_reward_func": 0.6666666666666666,
-      "rewards/question_recreation_reward_func": 0.38282742351293564,
+      "rewards/final_correctness_reward_func": 0.0,
+      "rewards/question_recreation_reward_func": 0.47715888544917107,
       "rewards/soft_format_reward_func": 0.0,
-      "rewards/strict_format_reward_func": 0.0,
-      "rewards/xmlcount_reward_func": 0.14445833986004195,
+      "rewards/strict_format_reward_func": 0.03125,
+      "rewards/xmlcount_reward_func": 0.3850937522947788,
       "step": 16
     },
     {
-      "completion_length": 315.71875,
-      "epoch": 4.571428571428571,
-      "grad_norm": 0.752004086971283,
-      "kl": 0.0008350697899004444,
-      "learning_rate": 3.013156219837776e-08,
+      "completion_length": 274.03125,
+      "epoch": 3.6,
+      "grad_norm": 0.6539471745491028,
+      "kl": 0.0008241815667133778,
+      "learning_rate": 1.2500000000000005e-07,
       "loss": 0.0,
-      "reward": 2.703938379883766,
-      "reward_std": 2.4889378771185875,
-      "rewards/concensus_correctness_reward_func": 0.9213124964153394,
-      "rewards/consensus_reward_func": 0.5,
+      "reward": 1.5772868506610394,
+      "reward_std": 1.0867214687168598,
+      "rewards/concensus_correctness_reward_func": 0.109375,
+      "rewards/consensus_reward_func": 0.125,
       "rewards/cumulative_reward_2": 0.0,
       "rewards/final_correctness_reward_func": 0.5625,
-      "rewards/question_recreation_reward_func": 0.5067508853971958,
+      "rewards/question_recreation_reward_func": 0.3505368586629629,
       "rewards/soft_format_reward_func": 0.0,
-      "rewards/strict_format_reward_func": 0.0,
-      "rewards/xmlcount_reward_func": 0.21337500168010592,
+      "rewards/strict_format_reward_func": 0.015625,
+      "rewards/xmlcount_reward_func": 0.41424999153241515,
       "step": 18
     },
     {
-      "completion_length": 289.0833333333333,
-      "epoch": 5.0,
-      "grad_norm": 0.3535054624080658,
-      "kl": 0.0008812252393302819,
-      "learning_rate": 3.4096741493194193e-09,
+      "completion_length": 296.59375,
+      "epoch": 4.0,
+      "grad_norm": 0.5222050547599792,
+      "kl": 0.0008954378354246728,
+      "learning_rate": 7.322330470336313e-08,
       "loss": 0.0,
-      "reward": 1.0508066192269325,
-      "reward_std": 0.8113361944754919,
-      "rewards/concensus_correctness_reward_func": 0.019499999471008778,
-      "rewards/consensus_reward_func": 0.16666666666666666,
+      "reward": 1.7973168343305588,
+      "reward_std": 1.314620640128851,
+      "rewards/concensus_correctness_reward_func": 0.07456249743700027,
+      "rewards/consensus_reward_func": 0.1875,
       "rewards/cumulative_reward_2": 0.0,
-      "rewards/final_correctness_reward_func": 0.16666666666666666,
-      "rewards/question_recreation_reward_func": 0.3977649410565694,
+      "rewards/final_correctness_reward_func": 0.625,
+      "rewards/question_recreation_reward_func": 0.6552544049918652,
       "rewards/soft_format_reward_func": 0.0,
-      "rewards/strict_format_reward_func": 0.0,
-      "rewards/xmlcount_reward_func": 0.3002083438138167,
+      "rewards/strict_format_reward_func": 0.015625,
+      "rewards/xmlcount_reward_func": 0.2393750031478703,
       "step": 20
     },
     {
+      "completion_length": 270.375,
+      "epoch": 4.4,
+      "grad_norm": 0.41121411323547363,
+      "kl": 0.0008414066614932381,
+      "learning_rate": 3.349364905389032e-08,
+      "loss": 0.0,
+      "reward": 1.6854848377406597,
+      "reward_std": 0.9369991458952427,
+      "rewards/concensus_correctness_reward_func": 0.015625,
+      "rewards/consensus_reward_func": 0.25,
+      "rewards/cumulative_reward_2": 0.0,
+      "rewards/final_correctness_reward_func": 0.625,
+      "rewards/question_recreation_reward_func": 0.4195785839110613,
+      "rewards/soft_format_reward_func": 0.0,
+      "rewards/strict_format_reward_func": 0.0,
+      "rewards/xmlcount_reward_func": 0.3752812519669533,
+      "step": 22
+    },
+    {
+      "completion_length": 311.40625,
+      "epoch": 4.8,
+      "grad_norm": 0.4889400601387024,
+      "kl": 0.0010594725827104412,
+      "learning_rate": 8.518543427732949e-09,
+      "loss": 0.0,
+      "reward": 1.1538150012493134,
+      "reward_std": 0.8227499965578318,
+      "rewards/concensus_correctness_reward_func": 0.03099999949336052,
+      "rewards/consensus_reward_func": 0.375,
+      "rewards/cumulative_reward_2": 0.0,
+      "rewards/final_correctness_reward_func": 0.0625,
+      "rewards/question_recreation_reward_func": 0.4522524643689394,
+      "rewards/soft_format_reward_func": 0.0,
+      "rewards/strict_format_reward_func": 0.0,
+      "rewards/xmlcount_reward_func": 0.23306249594315886,
+      "step": 24
+    },
+    {
+      "completion_length": 318.4375,
       "epoch": 5.0,
-      "step": 20,
+      "kl": 0.0008752793655730784,
+      "reward": 3.1140042692422867,
+      "reward_std": 3.7438226342201233,
+      "rewards/concensus_correctness_reward_func": 1.46875,
+      "rewards/consensus_reward_func": 0.125,
+      "rewards/cumulative_reward_2": 0.0,
+      "rewards/final_correctness_reward_func": 0.75,
+      "rewards/question_recreation_reward_func": 0.4615042470395565,
+      "rewards/soft_format_reward_func": 0.0,
+      "rewards/strict_format_reward_func": 0.0,
+      "rewards/xmlcount_reward_func": 0.3087499998509884,
+      "step": 25,
       "total_flos": 0.0,
-      "train_loss": 7.424532896038727e-07,
-      "train_runtime": 344.1703,
-      "train_samples_per_second": 0.93,
-      "train_steps_per_second": 0.058
+      "train_loss": 1.0554939626672422e-06,
+      "train_runtime": 1935.7748,
+      "train_samples_per_second": 0.207,
+      "train_steps_per_second": 0.013
     }
   ],
   "logging_steps": 2,
-  "max_steps": 20,
+  "max_steps": 25,
   "num_input_tokens_seen": 0,
-  "num_train_epochs": 7,
+  "num_train_epochs": 5,
   "save_steps": 25,
   "stateful_callbacks": {
     "TrainerControl": {

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9c91090e7d9ee3274d13ccbc81952e2cccb8091ff6c3aaa8b215a1ccce4c266b
+oid sha256:c45d95aa6abf8391b5945b8c407039aaf2b430adfcaa8a3c388622de5efa1b67
 size 6008